Research map / Detection and segmentation / CNN features
Ada-Segment: Automated Multi-loss Adaptation for Panoptic Segmentation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use adaptive multi-loss weighting instead of equal loss weights | Panoptic segmentation model · COCO val | PQ: 41.0 -> 43.7 | Helped |
| Use adaptive multi-loss weighting instead of coarse grid-searched weights | Panoptic segmentation model · COCO val | PQ: 42.1 -> 43.7 | Helped |
| Remove the weight controller and rely on evolution-strategy weight candidates (ablation) | Ada-Segment · COCO val | PQ: 43.2 -> 41.1 | Hurt |
| Add policy-ensemble training after exploration (ablation) | Ada-Segment · COCO val | PQ: 42.6 -> 43.2; additional 0.6% | Helped |
| Use fixed final exploration weights instead of dynamic adaptation (ablation) | Ada-Segment · COCO val | PQ: 43.7 -> 42.2 | Hurt |
| Use static weights predicted from initial losses instead of dynamic adaptation (ablation) | Ada-Segment · COCO val | PQ: 43.7 -> 42.6 | Hurt |
| Use the final policy dynamically without ensembling policy states (ablation) | Ada-Segment · COCO val | PQ: 43.7 -> 43.1 | Hurt |
| Apply Ada-Segment weighting to Panoptic FPN instead of equal weights | Panoptic FPN · COCO val | PQ: 38.1 -> 39.9; 1.8% | Helped |
| Transfer the policy trained on ADE20K proxy data to COCO (ablation) | Ada-Segment policy · COCO | PQ: 42.7 | Helped |
| Transfer the policy trained on COCO proxy data to ADE20K (ablation) | Ada-Segment policy · ADE20K | PQ: 31.6 -> 32.0 | Helped |
| Use the policy learned with a ResNet-50 training backbone on a ResNet-101 model (ablation) | Ada-Segment policy · COCO | PQ: 45.1 | Helped |
| Use only local reward, omitting improvement from the previous checkpoint (ablation) | Ada-Segment controller · COCO val | PQ: 43.7 -> 42.6 | Hurt |
Closest papers by meaning
- Cross-View Regularization for Domain Adaptive Panoptic Segmentation (CVPR 2021 · similarity 0.90)
- Auto-Panoptic: Cooperative Multi-Component Architecture Search for Panoptic Segmentation (NeurIPS 2020 · similarity 0.89)
- ADAPT: Attentive Self-Distillation and Dual-Decoder Prediction Fusion for Continual Panoptic Segmentation (ICLR 2025 · similarity 0.89)
- UniDAformer: Unified Domain Adaptive Panoptic Segmentation Transformer via Hierarchical Mask Calibration (CVPR 2023 · similarity 0.89)
- Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation (CVPR 2020 · similarity 0.88)
- Combinatorial Optimization for Panoptic Segmentation: A Fully Differentiable Approach (NeurIPS 2021 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.